The document presents an image enhancement technique that leverages the Retinex theory applied in the HSV color space with saturation correction. It discusses various methods, including multi-scale Retinex (MSR), and compares the processing times of the proposed approach with existing methods. Results indicate improved performance in terms of speed without compromising image quality.

1. Image Enhancement Technique Using Retinex of
HSV Color Space and Saturation Correction
ISL lab Seminar
Han-Sol Kang
26 June 2017

2. 5 October 2018
2
Contents
Introduction Retinex Proposed method Results

3. Introduction
Image quality
5 October 2018
3
Light source
Camera
Object

4. Introduction
Solutions
5 October 2018
4
Image
enhancement
Histogram
Eq.
RetinexGamma
correction
I
r
I
r
Output HistogramInput Histogram
/1
IS 

5. Retinex
Concept
5 October 2018
5
Light source
Camera
Object
Reflectance
),(),(),( yxLyxRyxI 
Reduce the effects of light source
x

6. Retinex
SSR
5 October 2018
6
),(),(),( yxLyxRyxI iii 
),(/),(),( yxLyxIyxR iii 
 ),(/),(log),( yxLyxIyxR iii 
Weber-Fechner’s law
Stimulus
Strength
   ),(log),(log),( yxLyxIyxR iii 
   ),(*),(log),(log),( yxIyxFyxIyxR iii 
Surround function
• Surround function by E.Land’86 (Inverse square spatial surround)
• Surround function by Hurlbert’89(Gaussian)

7. Retinex
MSR & MSRCR
5 October 2018
7



N
n
n
N
n
nnMSR WyxRWyxR ii
11
1),,(),(
   ),(*),(log),(log),( yxIyxFyxIyxR iiSSRi

]),(),([),( byxRyxCGyxR ii MSRiMSRCR 
Since the channels are operated independently, the colors are unnatural.
Gain-Offset

8. Proposed method
Concept
5 October 2018
8
Input(R,G,B) Output(R,G,B)
Input(V) Output(V)RGB color space HSV color space
Conventional Retinex
MSRMS
*
*
*
*

9. Proposed method
Concept
5 October 2018
9
Input(V) Output(V)
MSRMS
Input(S) Output(S)Output(V)
*
*
*

10. Proposed method
MSR for V channel
5 October 2018
10
255
),,max(
),,max(
),,min(),,max(
))(()(
)]()[(5.0
cos
2
1
BGR
V
BGR
BGRBGR
S
BGBRGR
BRGR
H













 
V channel
Surround function
*
V channel
-

11. Proposed method
Saturation correction(Modified Saturation)
5 October 2018
11
Input(S) Output(S)Output(V)
**
),()1(),(*
yxSyxS 







)),(),((0
)),(),((255/)),(),((
*
**
yxVyxV
yxVyxVyxVyxV


12. Proposed method
Contrast enhancement (HE)
5 October 2018
12







))),(((),(
))),((()),((
),( **
**
**


yxVStdyxV
yxVStdyxVHE
yxV

13. Results
5 October 2018
13

14. Results
5 October 2018
14

15. Results
5 October 2018
15

16. Results
5 October 2018
16
Resolution
Algorithm
MSR MSRCR MSRMS
600x394 0.523 sec 0.536 sec 0.186 sec
1280x720 2.642 sec 2.674 sec 0.974 sec
1920x1080 8.672 sec 8.842 sec 3.528 sec
Table 1. Comparison of processing time of the proposed method with those of MSR and MSRCR

17. Conclusions
5 October 2018
17

18. Q & A